Update app.py

app.py

@@ -1,79 +1,173 @@
-diff --git a/app.py b/app.py
-index 0000000..1111111 100644
---- a/app.py
-+++ b/app.py
-@@ -1,16 +1,28 @@
- import gradio as gr
- import torch
- from PIL import Image
- from transformers import AutoTokenizer, AutoModelForCausalLM
--import spaces
+import gradio as gr
+import torch
+from PIL import Image
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import spaces
 
- # Model configuration
- MID = "apple/FastVLM-0.5B"
- IMAGE_TOKEN_INDEX = -200
+# Model configuration
+MID = "apple/FastVLM-0.5B"
+IMAGE_TOKEN_INDEX = -200
 
- # Load model and tokenizer (will be loaded on first GPU allocation)
- tok = None
- model = None
- def load_model():
-     global tok, model
-     if tok is None or model is None:
-         print("Loading model...")
-         tok = AutoTokenizer.from_pretrained(MID, trust_remote_code=True)
--        model = AutoModelForCausalLM.from_pretrained(
--            MID,
--            torch_dtype=torch.float16,
--            device_map="cuda",
--            trust_remote_code=True,
--        )
-+        # ---- CPU-first, with dynamic fallback if CUDA exists ----
-+        use_cuda = torch.cuda.is_available()
-+        device_map = "cuda" if use_cuda else "cpu"
-+        # float16 is great on GPU, but unsafe on CPU; use float32 on CPU
-+        dtype = torch.float16 if use_cuda else torch.float32
-+
-+        model = AutoModelForCausalLM.from_pretrained(
-+            MID,
-+            torch_dtype=dtype,
-+            device_map=device_map,
-+            trust_remote_code=True,
-+        )
-         print("Model loaded successfully!")
-     return tok, model
--
-[email protected](duration=60)
-+
-+# Removed GPU decorator so CPU Spaces don't request a GPU
- def caption_image(image, custom_prompt=None):
-@@ -66,16 +78,23 @@ def caption_image(image, custom_prompt=None):
-     # Insert IMAGE token id at placeholder position
--    img_tok = torch.tensor([[IMAGE_TOKEN_INDEX]], dtype=pre_ids.dtype)
--    input_ids = torch.cat([pre_ids, img_tok, post_ids], dim=1).to(model.device)
--    attention_mask = torch.ones_like(input_ids, device=model.device)
-+    # Derive device/dtype from model parameters (robust on CPU or GPU)
-+    model_device = next(model.parameters()).device
-+    model_dtype = next(model.parameters()).dtype
-+
-+    img_tok = torch.tensor([[IMAGE_TOKEN_INDEX]], dtype=pre_ids.dtype, device=model_device)
-+    input_ids = torch.cat([pre_ids.to(model_device), img_tok, post_ids.to(model_device)], dim=1)
-+    attention_mask = torch.ones_like(input_ids, device=model_device)
+# Load model and tokenizer (will be loaded on first GPU allocation)
+tok = None
+model = None
 
-     # Preprocess image using model's vision tower
-     px = model.get_vision_tower().image_processor(
-         images=image, return_tensors="pt"
-     )["pixel_values"]
--    px = px.to(model.device, dtype=model.dtype)
-+    px = px.to(model_device, dtype=model_dtype)
+def load_model():
+    global tok, model
+    if tok is None or model is None:
+        print("Loading model...")
+        tok = AutoTokenizer.from_pretrained(MID, trust_remote_code=True)
+        model = AutoModelForCausalLM.from_pretrained(
+            MID,
+            torch_dtype=torch.float16,
+            device_map="cuda",
+            trust_remote_code=True,
+        )
+        print("Model loaded successfully!")
+    return tok, model
 
-     # Generate caption
-     with torch.no_grad():
-         out = model.generate(
-             inputs=input_ids,
-             attention_mask=attention_mask,
-             images=px,
-             max_new_tokens=128,
-             do_sample=False,  # Deterministic generation
--            temperature=1.0,
-+            # temperature is ignored when do_sample=False
-         )
+@spaces.GPU(duration=60)
+def caption_image(image, custom_prompt=None):
+    """
+    Generate a caption for the input image.
+    
+    Args:
+        image: PIL Image from Gradio
+        custom_prompt: Optional custom prompt to use instead of default
+    
+    Returns:
+        Generated caption text
+    """
+    if image is None:
+        return "Please upload an image first."
+    
+    try:
+        # Load model if not already loaded
+        tok, model = load_model()
+        # Convert image to RGB if needed
+        if image.mode != "RGB":
+            image = image.convert("RGB")
+        
+        # Use custom prompt or default
+        prompt = custom_prompt if custom_prompt else "Describe this image in detail."
+        
+        # Build chat message
+        messages = [
+            {"role": "user", "content": f"<image>\n{prompt}"}
+        ]
+        
+        # Render to string to place <image> token correctly
+        rendered = tok.apply_chat_template(
+            messages, add_generation_prompt=True, tokenize=False
+        )
+        
+        # Split at image token
+        pre, post = rendered.split("<image>", 1)
+        
+        # Tokenize text around the image token
+        pre_ids = tok(pre, return_tensors="pt", add_special_tokens=False).input_ids
+        post_ids = tok(post, return_tensors="pt", add_special_tokens=False).input_ids
+        
+        # Insert IMAGE token id at placeholder position
+        img_tok = torch.tensor([[IMAGE_TOKEN_INDEX]], dtype=pre_ids.dtype)
+        input_ids = torch.cat([pre_ids, img_tok, post_ids], dim=1).to(model.device)
+        attention_mask = torch.ones_like(input_ids, device=model.device)
+        
+        # Preprocess image using model's vision tower
+        px = model.get_vision_tower().image_processor(
+            images=image, return_tensors="pt"
+        )["pixel_values"]
+        px = px.to(model.device, dtype=model.dtype)
+        
+        # Generate caption
+        with torch.no_grad():
+            out = model.generate(
+                inputs=input_ids,
+                attention_mask=attention_mask,
+                images=px,
+                max_new_tokens=128,
+                do_sample=False,  # Deterministic generation
+                temperature=1.0,
+            )
+        
+        # Decode and return the generated text
+        generated_text = tok.decode(out[0], skip_special_tokens=True)
+        
+        # Extract only the assistant's response
+        if "assistant" in generated_text:
+            response = generated_text.split("assistant")[-1].strip()
+        else:
+            response = generated_text
+            
+        return response
+        
+    except Exception as e:
+        return f"Error generating caption: {str(e)}"
+
+# Create Gradio interface
+with gr.Blocks(title="FastVLM Image Captioning") as demo:
+    gr.Markdown(
+        """
+        # 🖼️ FastVLM Image Captioning
+        
+        Upload an image to generate a detailed caption using Apple's FastVLM-0.5B model.
+        You can use the default prompt or provide your own custom prompt.
+        """
+    )
+    
+    with gr.Row():
+        with gr.Column():
+            image_input = gr.Image(
+                type="pil",
+                label="Upload Image",
+                elem_id="image-upload"
+            )
+            
+            custom_prompt = gr.Textbox(
+                label="Custom Prompt (Optional)",
+                placeholder="Leave empty for default: 'Describe this image in detail.'",
+                lines=2
+            )
+            
+            with gr.Row():
+                clear_btn = gr.ClearButton([image_input, custom_prompt])
+                generate_btn = gr.Button("Generate Caption", variant="primary")
+        
+        with gr.Column():
+            output = gr.Textbox(
+                label="Generated Caption",
+                lines=8,
+                max_lines=15,
+                show_copy_button=True
+            )
+    
+    # Event handlers
+    generate_btn.click(
+        fn=caption_image,
+        inputs=[image_input, custom_prompt],
+        outputs=output
+    )
+    
+    # Also generate on image upload if no custom prompt
+    image_input.change(
+        fn=lambda img, prompt: caption_image(img, prompt) if img is not None and not prompt else None,
+        inputs=[image_input, custom_prompt],
+        outputs=output
+    )
+    
+    gr.Markdown(
+        """
+        ---
+        **Model:** [apple/FastVLM-0.5B](https://huggingface.co/apple/FastVLM-0.5B)
+        
+        **Note:** This Space uses ZeroGPU for dynamic GPU allocation.
+        """
+    )
+
+if __name__ == "__main__":
+    demo.launch(
+        share=False,
+        show_error=True,
+        server_name="0.0.0.0",
+        server_port=7860
+    )